| Περίληψη: | Object detection is a computer technology field related to image processing and computer vision. Object detection deals with locating and identifying an object belonging to a specific class in an image or a video. In recent years, object detection in real time has been part of many research and industrial applications, especially after the breakout popularity of Convolutional Neural Networks. Some of those applications are self-driving cars, medical imaging interpretation, anomaly detection in scenes, face detection etc.
In this thesis, we use and compare many different methods and approaches, in order to detect vehicles from a given video in real time. After a vehicle is detected, we extract the license plate number. The above methods are used on an embedded system. We train object detection and OCR models offline and the use them on the embedded system of choice: The Raspberry Pi.
For the car detection portion of this thesis, several different Convolutional Neural Networks have been used. Training data is a mosaic of different datasets, in order to create a better model. Some pre-trained models are also used. All methods have one thing in common: they not only detect the vehicle, but its coordinates in the image. After a vehicle is detected, an edge detection algorithm is used to detect the position of the license plate and crop the image. Lastly, the cropped image is used as input to an OCR algorithm, which extracts the license plate number.
All the above is applied in each frame of a video. We encounter each frame as an independent image, for simplicity.
|
|---|
